Pattern-Aware Diffusion Synthesis of fMRI/dMRI with Tissue and Microstructural Refinement

TL;DR

This work tackles the challenge of missing MRI modalities by proposing Pattern-aware Diffusion Synthesis (PDS), a bidirectional, 3D cross-modal diffusion framework that injects disease semantics into synthesis. It combines a Pattern-aware Dual-modal Diffusion Model (PDM) with Tissue Refinement (TR) and Microstructure Refinement (MR) modules, trained through a two-stage loss that first establishes semantic-consistent generation and then sharpens anatomical and microstructural details. Across three datasets (ADNI, OASIS-3, and a hospital cohort), PDS achieves state-of-the-art fidelity (PSNR/SSIM improvements) and demonstrates clinically meaningful diagnostic performance in NC/MCI/AD classification. The approach offers a practical route to cost-effective, multimodal neuroimaging with potential for early detection and personalized management of neurodegenerative diseases, with code available publicly.

Abstract

Magnetic resonance imaging (MRI), especially functional MRI (fMRI) and diffusion MRI (dMRI), is essential for studying neurodegenerative diseases. However, missing modalities pose a major barrier to their clinical use. Although GAN- and diffusion model-based approaches have shown some promise in modality completion, they remain limited in fMRI-dMRI synthesis due to (1) significant BOLD vs. diffusion-weighted signal differences between fMRI and dMRI in time/gradient axis, and (2) inadequate integration of disease-related neuroanatomical patterns during generation. To address these challenges, we propose PDS, introducing two key innovations: (1) a pattern-aware dual-modal 3D diffusion framework for cross-modality learning, and (2) a tissue refinement network integrated with a efficient microstructure refinement to maintain structural fidelity and fine details. Evaluated on OASIS-3, ADNI, and in-house datasets, our method achieves state-of-the-art results, with PSNR/SSIM scores of 29.83 dB/90.84\% for fMRI synthesis (+1.54 dB/+4.12\% over baselines) and 30.00 dB/77.55\% for dMRI synthesis (+1.02 dB/+2.2\%). In clinical validation, the synthesized data show strong diagnostic performance, achieving 67.92\%/66.02\%/64.15\% accuracy (NC vs. MCI vs. AD) in hybrid real-synthetic experiments. Code is available in \href{https://github.com/SXR3015/PDS}{PDS GitHub Repository}

Figures (8)

• Figure 1: We divide the uneven time axis into N segments to align the temporal/gradient dimensions between fMRI and dMRI. Then, to implement the cross-modal fMRI/DTI synthesis, we employ the pattern-aware dual-modal diffusion model (PDM), tissue refinement (TR), and microstructure refinement (MR) to improve the image quality and clinical applicability.
• Figure 2: Pattern-aware Diffusion Model. Our diffusion model incorporates a noise estimator (NE) for initial fMRI and dMRI generation through progressive denoising. A pattern-aware (DA) module ensures semantic consistency (pattern at the brain disease-related regions of atlas ) across noisy image and noise. Tissue and microstructure details are refined through a dedicated projection network and a microstructure refinement loss $\mathcal{L}_{mic}$, where $\gamma_t = \sqrt{1 - \bar{\alpha}_t}$.
• Figure 3: The details of pattern-aware module.
• Figure 4: Qualitative comparison across axial, sagittal, and coronal projections demonstrates our method's superior synthesis capabilities, achieving enhanced anatomical precision and improved tissue microstructure alignment with ground-truth.
• Figure 5: Visualization comparison pattern results of roi signals (fMRI/dRMI) in different kinds of cognitive status